Evapotranspiration Estimates from Eddy Covariance Towers and Hydrologic Modeling in Managed Forests in Northern Wisconsin, USA

agricultural and forest meteorology 148 (2008) 257–267

available at www.sciencedirect.com

journal homepage: www.elsevier.com/locate/agrformet

Evapotranspiration estimates from eddy covariance towers and hydrologic modeling in managed forests in Northern Wisconsin, USA

G. Sun a,*, A. Noormets b, J. Chen b, S.G. McNulty a a Southern Global Change Program, USDA Forest Service, Raleigh, NC 27606, United States b Department of Environmental Sciences, University of Toledo, Ohio 43606, United States article info abstract

Keywords: Direct measurement of ecosystem evapotranspiration by the eddy covariance method and Evapotranspiration simulation modeling were employed to quantify the growing season (May–October) evapo- Eddy covariance transpiration (ET) of eight forest ecosystems representing a management gradient in domi- MIKE SHE modeling nant forest types and age classes in the Upper Great Lakes Region from 2002 to 2003. We Management measured net exchange of water vapor fluxes in a 63-year-old mature hardwood (MHW) stand, Wisconsin a 60-year-old mature red pine (MRP) stand, a 3-year-old young hardwood (YHW) stand, a 17- year-old intermediate hardwood (IHW) stand, a young red pine (YRP age 8) stand, an intermediate red pine (IRP age 21) stand, and two pine barren ecosystems burned 12 years (PB1) and 2 years (PB2) ago. Field data suggested that there were no significant differences in growing season (June–September) ET/precipitation ratio among all ecosystems in 2002. However, PB2 had significantly lower ET/precipitation than those of other ecosystems in 2003. The ratios were much higher for all ecosystems, up to 0.90 for IHW, during the peak summer months (June–July). PB2 was the lowest (0.64) during that period. Stand leaf area index alone did not explain ecosystem ET at the landscape scale. Seasonal ET values measured by the eddy covariance method were significantly lower than those simulated with a process-based hydrologic model, MIKE SHE. Our integration approach combined with field measurements and simulation modeling proved to be useful in providing a full picture of the effects of forest cover type change on landscape scale water balance at multiple temporal scales. The ET procedure used in the MIKE SHE model needs improvement to fully account for the effects of vapor pressure deficit on tree transpiration. Seasonal distributions of ET coincided with precipitation in the growing season, when fluxes estimated by both field and models were the highest. The simulation model suggests that removal of conifer forests in the study region may reduce ET immediately by 113–30 mm/year or about 20%, but our field data suggests that ET can recover within 8–25 years from re-growth of hardwood forests. # 2007 Elsevier B.V. All rights reserved.

1. Introduction and elsewhere in the United States (Ice and Stednick, 2004)and around the world (Andreassian, 2004) have shown that forest Century-long studies of the forest–water relations in the Lake management plays a great role in regulating the hydrological States region (Verry, 1986; Mackay et al., 2002; Ewers et al., 2002) cycle, such as streamﬂow and evapotranspiration, at multiple

* Corresponding author. Tel.: +1 919 515 9498; fax: +1 919 513 2978. E-mail address: [email protected] (G. Sun). 0168-1923/$ – see front matter # 2007 Elsevier B.V. All rights reserved. doi:10.1016/j.agrformet.2007.08.010 258 agricultural and forest meteorology 148 (2008) 257–267

temporal and spatial scales by altering ecosystem water the sum of individual ecosystem functions and responses balances. However, hydrologic responses to land management (Chen et al., 2004). Studies that quantify and contrast water and cover change vary greatly because of the complex balances of multiple, dominant ecosystems of a landscape are interactions among climate, soil, and vegetation from indivi- critically important as a basis for modeling land–atmosphere dual tree to landscape scales (Andreassian, 2004; Ewers et al., interactions at broader scales. 2002; Mackay et al., 2002). At the watershed scale, ET represents the largest water flux The Midwest United States has experienced one of the next to precipitation, but it is the most challenging variable to largest land use changes in the world. It has been estimated measure at this scale due to the heterogeneity of the that 137 million acres of virgin forest in this region was landscape. In practice, ET is often calculated as the residual converted to agriculture and urban uses between the 1620s from precipitation and other water fluxes (e.g., runoff, change and 1990s (Verry, 2004). The remaining forests are highly in soil water storage) and is subject to large errors at a short fragmented. For example, the Chequamegon-Nicolet National temporal scale and in watersheds with poorly defined surface Forest in Northern Wisconsin, an area of approximately and subsurface watershed boundaries. The energy balance 325,000 ha, was logged, mainly for pine, from 1860 to 1920, method and eddy covariance technique provide alternative and the forest has been regrown for the past 85 years (Bresee measures of latent heat flux, equivalent to ET, and offer et al., 2004). The consequences of this large-scale forest promising estimates for closing the water balances of change on water balances have not been studied in detail. In ecosystems (Wilson et al., 2001; Law et al., 2002). Meanwhile, the Lake States, clearcutting upland hardwoods or conifers water flux measured at shorter temporal scales has been could increase annual streamflow by 90–200 mm/year (30– scaled up to construct annual water budgets at ecosystem 80% increase). Yet a similar forest management practice scale (Gholz and Clark, 2002; Thornton et al., 2002; Mackay would have minor hydrologic effect on peatlands (Verry, 1986). et al., 2002; Ewers et al., 2002). The ET data collected at eddy To empirically investigate changes in values of key hydro- flux towers have been widely used for validating traditional logical variables and identify potential biophysical drivers, hydrologic models to extrapolate findings at individual sites to process-based studies were conducted at tree species, stand, the regional scale (Wilson et al., 2001). and landscape levels in forests of Northern Wisconsin to This paper examines measured ET from several eddy flux directly measure sapflow (Ewers et al., 2002; Mackay et al., towers in forest ecosystems along a management gradient in 2002), and water and carbon fluxes using eddy flux towers the Chequamegon-Nicolet National Forest in Northern Wis- (Baldocchi et al., 1996). These studies reported more than consin and evaluated modeled ET using a biophysically based twofold differences in evapotranspiration (ET) among the integrated forest hydrologic model, MIKE SHE (DHI, 2004). We dominant forests in the region (i.e., northern hardwoods, red hypothesized that: (1) coniferous forests have higher ET than pine/Jack pine, aspen/fir, and forested wetlands), primarily hardwood forests because the former can lose water through due to differences in total sapwood area and tree hydraulic interception and transpiration during non-growing seasons; conductance and a much lesser extent to differences in total (2) the ET of mature forests is higher than that of young forests leaf area. These studies further suggested that forested because mature forests have more leaf area (i.e., ET is age- wetlands had significantly higher total ET than uplands, dependent) (Chen et al., 2002). Specifically, our study objec- calling for physically based hydrologic models for more tives were to: (1) compare ET changes over time in multiple accurately estimating soil evaporation water fluxes in the ecosystems; (2) evaluate the effect of forest age on ET; (3) region. Similar effects of forest changes on water balances simulate changes of ET using a hydrological model to expand were also reported for the Southern USA. Swank et al. (1988) our capability to predict ET for the dominant ecosystems of the found that clearcutting a southern hardwoods forest in the region. Appalachian Mountains in western North Carolina could result in a 200–400 mm/year increase in water yield, mostly because of reduction in ecosystem ET. In contrast, clearing of 2. Materials and methods the slash pine (Pinus elliottii) forests on flatwoods in north Florida had limited and short-term effects on ET, and thus 2.1. Study sites water yield, especially during wet periods (Gholz and Clark, 2002; Bliss and Comerford, 2002; Sun et al., 1998). The study sites (468300–468450N, 91820–918220W) were located in Recently, concern about global climate change has led the Chequamegon-Nicolet National Forest, in Ashland County researchers in forest hydrology to look more closely at in Northern Wisconsin, USA (Fig. 1). The landscape is generally connections between the water and nutrient cycles and the flat and homogeneous. The predominant soils are well- dynamics of energy and carbon cycles (McNulty et al., 1997; drained loamy tills with ground moraine, non-calcareous Sun et al., 2000; Schafer et al., 2002). Accumulated field data sandy loamy tills, and outwash sand. Sandy loams are found (Wilson et al., 2002) and meso-scale simulations (Werth and below 30 cm from the ground surface, and the groundwater Avissar, 2002) suggest that land surface characteristics at the table was at depths greater than 2 m in the uplands, but was at landscape scale affect water and energy partitioning and that the surface in wetland areas (Ewers et al., 2002). The climate is land–atmosphere feedbacks can be manifested at large scales humid-continental with average temperature ranging from (Pielke et al., 1998). Scaling up stand-level flux measurements 13 8C in January to 20 8C in July, and averaged annual to landscape scales (the bottom-up approach) remains a precipitation varies between 600 and 900 mm with 70% of this theoretical challenge because ecosystem functions and occurring during the May–October growing season. Annual responses to disturbances at the regional scale do not equal long-term ET in the region is 560 mm or 70% of annual agricultural and forest meteorology 148 (2008) 257–267 259

Fig. 1 – Eddy covariance tower locations in the Chequamegon-Nicolet National Forest in Northern Wisconsin, USA.

precipitation as estimated from long-term hydrology (1951– collected between 1 January 2002 and 31 December 2003. Net 2 1980) and climatic records (Gerbert et al., 1987; Daly et al., radiation (Rn,Wm ) above the stand canopy was measured 2000). The vegetation is dominated by second-growth hard- with a Q7.1 net radiometer (radiation and energy balance woods (40–45%) and conifer stands including red pine (Pinus systems, REBS, Seattle, WA, USA) and air temperature (Ta, 8C) resinosa) and jack pine (Pinus banksiana) plantations (35%), and and relative humidity (%) were measured above, within, and pine barrens (17%) (Bresee et al., 2004). below the canopy (1.8 m) using HMP45AC probes (Vaisala, We selected multiple ecosystems for this 2-year study to Finland). Soil temperature (8C) and matric potential (bars) were investigate how these ecosystems respond to land disturbance measured at a 30 cm depth using CS107 temperature probes and environmental variability at the landscape scale. The and CS257 gypsum moisture blocks (SCI, Logan, UT, USA). ecosystem types included mature hardwoods (MHW) domi- Precipitation (mm) was measured with a tipping bucket rain nated by aspen-birch (Populus grandidentata, P. tremuloides, gauge (SCI) located at a permanent weather station 8–25 km Betula papyrifera), mature red pine (MRP), naturally regenerated from the individual stands. The daily values were derived from young hardwoods (clearcut in 2000) (YHW), young red pine (8- these 30 min raw data. When daily air temperature and year old), intermediate hardwoods (17-year old), intermediate precipitation data were not available due to equipment failure, red pine IRP (21-year old), a natural pine barren (PB1), and a manual measurements at the University of Wisconsin Ash- burned pine barren (PB2). Pine barrens are a temperate land Agricultural Research Station were used. savanna community dominated by scattered jack pines, oaks (Quercus spp.), shrubby hazelnuts (Corylus spp.) and prairie 2.3. Net exchange of energy and water vapor willow (Salix humilis), and herbs (Bresee et al., 2004). Leaf area index (LAI) was estimated by hemispheric photography and Net exchange of energy and water vapor was measured the WinScanopy image analysis program (Regent Instru- continuously in five ecosystems (MHW, MRP, YHW, YRP and ments, Quebec, Canada). Forest community characteristics PB1) in 2002 and five ecosystems (MHW, MRP, IHW, IRP and and experimental and modeling designs for the two experi- PB2) in 2003, a total of eight different year types of ecosystems, mental years, 2002 and 2003, are presented in Table 1. by employing the eddy covariance method. Each system included a Li-7500 open-path infrared gas analyzer (IRGA, Li- 2.2. Micrometeorological measurements Cor), a CSAT3 three-dimensional sonic anemometer (CSI), a CR5000 datalogger (CSI), and a barometer (CS105, CSI). The Meteorological measurements at a 30 min resolution were 30 min mean latent heat flux was calculated after screening initiated in January 2002. Our data analysis included data the data for periods of precipitation, instrument or power 260 agricultural and forest meteorology 148 (2008) 257–267

Table 1 – Stand characteristics and experimental design of seven forest ecosystems across a management gradient in the Chequamegon-Nicolet National Forest, Wisconsin, USA Ecosystem Age (years) Canopy Basal area Maximum Measurement/modeling in 2002 cover (%) (m2 ha1) leaf area index periods

Mature hardwood (MHW) 65 97 29.1 4.0 2002, 2003 simulation applied Mature red pine (MRP) 63 70 13.5 2.8 2002, 2003 simulation applied Young hardwood (YHW) 3 2 1.5 0.5 2002 simulation applied Young red pine (YRP) 8 17 4.6 0.52–0.93 2002 Pine barrens (PB1) 12 4 0.5 0.2 2002 simulation applied Pine barrens (burned) (PB2) 2 0 0.0 0.05 2003 Intermediate hardwood (IHW) 17 N/A N/A 3 2003 Intermediate pine (IRP) 21 N/A N/A 3 2003 failure, low turbulence conditions, and out-of-range records. variable (Lu et al., 2005). In this study, we computed PET Turbulent exchange of water vapor and energy were calcu- externally by Hamon’s method (Federer and Lash, 1978; Lu lated according to Leuning (2004) using two-axis coordinate et al., 2005) before running the hydrologic model. Daily rotation. Sonic temperature was corrected according to Hamon’s PET was computed as a function of measured air Schotanus and Nieuwstadt (1983), and the fluxes were temperature and calculated daytime length using site location corrected for fluctuations in air density using the Webb– (latitude) and date (Julian Day). Pearman–Leuning expression (Webb et al., 1980). Data quality was assessed using the stationarity and stability criteria PET ¼ 0:1651DVdk (1) (Foken and Wichura, 1996), along with range and variance 1 checks. Energy balance closure was 56–68% (Nan Lv unpub- where PET is potential evapotranspiration (mm day ), D lished personal communication) and was not used for data the time from sunrise to sunset in multiples of 12 h, computed screening. Wind characteristics and turbulent fluxes did not as a function of date, latitude, and slope and aspect of the 3 show detectable directional variation, and therefore fluxes watershed; Vd the saturated vapor density (g m ) at the daily from behind the sensor were not explicitly removed, but mean temperature, T (8C) = 216.7Vs/(T + 273.3); Vs the satu- subjected to the standard screening described above. The rated vapor pressure (mbar) = 6.108 exp[17.26939T/ 30 min average of latent heat was further summed to a daily (T + 237.3)]; and k is the correction coefficient (1.1) to adjust scale. Latent heat fluxes for periods when the sensors were PET calculated using Hamon’s method to measured values. wet were rejected. We also discarded some daily latent heat Water loss through canopy interception (Ec) was modeled data when the values were abnormally higher than those of as a function of daily PET and daily leaf area index LAI (denoted net radiation. Daily ET in mm day1 was derived by dividing as L below): the LE by the vaporization constant.

Ec ¼ minðImax; PETÞ (2) 2.4. Water balance modeling

Imax ¼ CintL (3) We employed the physical forest hydrologic model MIKE SHE

(Abbott et al., 1986a,b; DHI, 2004; Tague et al., 2004; Im et al., where Imax is maximum canopy interception storage, which

2004) to estimate daily ET for 2002 and 2003 for three major must be filled before stemflow occurs. Cint is an interception contrasting forest ecosystems where field data over the 2- coefficient with a typical value of 0.05 mm. year-study period were available. These ecosystems were Plant transpiration (T) was described as a function of PET, L, mature red pine (MRP), mature hardwoods (MHW), and two root distribution, and soil moisture content in the rooting pine barrens with different burn history (PB1 and PB2). The zone: MIKE SHE model simulates the full hydrologic cycle char- acteristic of a forest ecosystem, including ET (i.e., sum of T ¼ f 1ðLÞ f 2ðuÞRPET (4) canopy interception, plan transpiration, and soil evaporation) and vertical soil water movement in the unsaturated soil zone where f1 and f2, are empirical functions of LAI, and soil moist- to the groundwater. A brief description of the modeling ure content (u) in the rooting zone and PET, respectively. R is a procedures and mathematical formulation are provided root distribution function that allows water to be extracted at below. Further details can be found in the model use manual varying soil depths over time. (DHI, 2004) and associated publications (Abbott et al., 1986a,b; Soil evaporation, as the last ET component, was simulated

Tague et al., 2004; Im et al., 2004). as a function of PET, L, soil moisture, and residual energy

One of the important components of the MIKE SHE model is availability (PET T): the ET submodel that interacts with atmospheric demand, soil moisture content, and groundwater table dynamics. As in Es ¼ PET f 3ðuÞþðPET T PET f 3ðuÞÞ f 4ðuÞð1 f 1ðLÞÞ (5) most physically based hydrologic models, total ET flux is controlled by potential evapotranspiration (PET) which is where f3 and f4 are empirical functions of soil moisture content. defined as maximum ecosystem ET under no-water-stress We assumed a flat land surface, no surface flow, and no net conditions. Several models are available to estimate this lateral groundwater flow in each of the forest ecosystems. We agricultural and forest meteorology 148 (2008) 257–267 261

used generalized soil and vegetation parameters for the study sites, but applied the measured daily climatic variables (air temperature and rainfall) in this study. The default parameters for the soil moisture release curves and hydraulic conductivity for sandy loam soils were used. The model was not calibrated because we did not intend to fit the model to the specific site conditions, but rather to estimate the potential errors in modeling by comparing simulation results with the eddy flux measurements. We hope to use the information gained to predict ET for other ecosystems in the region. In the simulation runs we focused on three ecosystems that represent a mature hardwood forest (Max LAI = 4.0), a mature red pine forest (Max LAI = 2.8), and PB1 and PB2 systems with low LAI (Max LAI = 0.5 and 0.05, respectively). Model input and Fig. 2 – Monthly precipitation, potential output (Table 2) are provided to give a general overview of evapotranspiration, and air temperature patterns during model requirements and functions. Detailed model descrip- 2002 and 2003 at the study site in North Wisconsin. tion, structure, and algorithms are found in the MIKE SHE user’s guide (DHI, 2004) and Abbott et al. (1986a,b). Analysis of variance (ANOVA) was performed to examine the differences among ecosystems. periods when the data were most complete for both years. The ANOVA at the daily temporal scale indicated that there were no significant ecosystem-to-ecosystem differences ( p < 0.05) 3. Results in ET among the five ecosystems in 2002. However, there were significant ecosystem-to-ecosystem differences ( p < 0.01) in Although forest evapotranspiration is relatively stable in 2003. Daily ET at the PB2 site was significantly lower than that comparison with other components of the water balance, it at the other four sites (MHW, MRP, IHW and IRP) (Fig. 3, is closely related to soil water availability as determined by Table 3). Values of ET for the other four ecosystems did not precipitation, runoff, and atmospheric demand. The monthly differ significantly ( p = 0.05). Linear regression analysis sug- rainfall distribution in 2002 was similar to the 30-year average gested that LAI was not significantly correlated with averaged except in October, which was wetter than average (Fig. 2). The daily ET of the five ecosystems in 2002 (R2 = 0.0). However, the year 2003 was drier than 2002 and the long-term average correlation between LAI and measured ET for the five except during May, September, and October. Because the soil ecosystems was significant (R2 = 0.96) in 2003 although there infiltration capacity was high, our study sites did not have any were no significant correlations among the four ecosystems noticeable surface runoff. However, slow internal subsurface (MRP, MHW, IHW and IRP). The correlation between ET and LAI drainage was expected to occur beneath the low-relief was heavily influenced by the PB1 and PB2 sites. Without data landscape. for these two sites, LAI was not effective in explaining the In general, daily ET for each of the ecosystem varied greatly variation in ET among ecosystems. during the growing seasons (May–October) (Fig. 3). The large Two examples of diurnal variation in dry canopy ET for day-to-day variability (>3–5-fold) in the summer months 2002 and 2003 are given in Fig. 4 to show how the ecosystems suggests a complex control on the ET processes. ET peaked differed on an hourly basis. For example, it appeared that red much earlier and higher in 2003 (Mid-July) than in 2002 (end of pines (MRP and YRP) had higher ET than the hardwoods (MHW July) (Fig. 3). Because many ET data points for May and October and YHW) in September 2002, while daytime ET for PB1 were missing or unusable, the ET data for May and October differed little from that for the rest of the ecosystems at that were not used in the cross-site comparisons in this study. time (Fig. 4a). ET was much greater on 18 July 2003, than on 11 Instead, we focused on ET comparison for the June–September September 2002 (Fig. 4b). However, there were no differences

Table 2 – Major model inputs and outputs of the hydrologic model, MIKE SHE Model overview Variables and parameters

Input requirements Climate Daily precipitation, air temperature for estimating potential evapotranspiration calculated using Hamon’s method (Federer and Lash, 1978) Vegetation Effective rooting depth (60 cm); daily leaf area index (LAI) estimated from literature and ﬁeld observation Soils Hydraulic conductivity and soil moisture release curve for sandy loams (Van Genuchten et al., 1991)

Output Daily canopy interception, transpiration, soil evaporation, soil moisture content (%) at different depths, groundwater table depth, soil water drainage from the rooting zone 262 agricultural and forest meteorology 148 (2008) 257–267

Fig. 3 – Measured daily ET by the eddy covariance method for: (a) mature hardwood (MHW), mature red pine (MRP), Fig. 4 – Diurnal change of ET for: (a) 11 September 2002 and recently clearcut hardwoods (YHW), young red pine (YRP), (b) 16 July 2003. and pine barrens (PB1) during 2002; (b) mature hardwood (MHW), mature red pine (MRP), intermediate red pine (IRP), intermediate hardwoods (YHW), and burned pine barrens (PB2) during 2003. sites had a much higher ET than in 2002 (Fig. 5). Except in May and September, PB2 had the lowest monthly ET throughout 2003. The fact that July ET was notably higher in 2003 than in 2002 for all sites suggests a significant water stress in July 2002. in ET among MRP, IRP, and MHW on 18 July 2003. The largest Precipitation was 110 mm in each year, but the July precipita- difference in ET occurred between IHW and PB2 on 18 July tion was 30 mm below PET in 2002 and the air temperature was 2003. The rest of the ecosystems had similar ET on this 2 8C higher in 2002 than in 2003 (Fig. 2). Soil moisture particular day. The ET comparison for these two dates further measurements confirmed this observation. For example, suggested that the relative magnitude of ET fluxes among measured daily ET for the MHW ecosystem in July 2002 was ecosystems had large variability. 3.0 1.1 mm (mean 1 S.D.) and the soil matric potential at Comparisons of peak ET values for the two summer the 30 cm soil depth was 1.1 0.98 bars, while ET and soil months (June–July) which had almost complete data showed potential were measured as 4.4 2.8 mm and 0.42 0.34 bars, a different contrasting pattern. There were no significant respectively, in July 2003. Our hydrological simulation model differences in ET ( p = 0.05) among the five ecosystems with indeed suggested a severe soil moisture stress during the diverse tree LAI ranging from 0.5 (YHW) to 4.0 (MHW). ANOVA summer months for both years (see later discussion). In suggested that there were significant differences in daily ET contrast, measured ET in August 2002 was lower than that in ( p < 0.01) among the five ecosystems measured in 2003. PB2 2003 (Fig. 5), presumably because the PET was 14 mm lower had the lowest average daily ET (2.1 0.9 mm day1), and the and the mean temperature was 2 8C lower than that in 2003 ET values for the other four ecosystems did not differ (Fig. 2). The simulation model did not show apparent water significantly in 2003. stress in August for either year. Monthly (June–September) ET comparisons across ecosys- During the growing season from June to September, ET tems for 2002 showed that MRP had the highest ET during the accounted for 60% (PB1) to 70% (MRP) of precipitation in 2002, early growing season (May), but that ET for the MHW site was and for 45% (PB2) to 70% (MHW) in 2003. As expected, the ET/P the highest during the peak month of July (Fig. 5). Monthly ET ratios were higher in June and July than the growing season values for PB1, YHW, and YRP, which had low leaf areas, were average, because ET was highest during these months. ET/P found to be comparable to those for the mature systems ratios ranged from 62% (YHW) to 75% (MHW) in June–July in during June, but 10–24 mm lower than July ET for MHW. 2002 and 64% (PB2) to 90% (IHW) in 2003 (Table 3). ET was the Similar contrasting patterns were found for July 2003, when all lowest in stands with low leaf area (PB1, PB2 and YHW), but agricultural and forest meteorology 148 (2008) 257–267 263

Table 3 – Summary of daily evapotranspiation (ET) across the eight ecosystems during the growing seasons in 2002 and 2003 Year Ecosystems Daily ET ﬂux (mean 1 S.D.) Periodic N Signiﬁcancea (mm day1) and min–max total ET/P (days)

2002 (May–October) (P = 560 mm) MHW 1.6 1.3 (0.0–5.1) 0.52 171 A MRP 1.9 1.2 (0.0–4.8) 0.61 163 A YHW 1.6 0.9 (0.3–3.6) 0.51 169 A YRP 1.7 0.1 (0.0–4.9) 0.55 179 A PB1 1.7 1.1 (0.2–4.2) 0.50 178 A

2003 (May–October) (P = 608 mm) MHW 2.0 1.3 (0.0–5.8) 0.61 114 B MRP 1.9 1.1 (0.1–4.7) 0.49 121 B IHW 1.8 1.5 (0.0–6.1) 0.57 107 B IRP 2.0 1.2 (0.0–5.5) 0.53 110 B PB2 1.4 1.0 (0.0–4.2) 0.41 145 C

2002 (June–July) (P = 197 mm) MHW 2.4 1.1 (0.6–5.1) 0.75 61 D MRP 2.2 1.2 (0.3–4.8) 0.68 61 D YHW 2.0 0.9 (0.5–3.6) 0.62 61 D YRP 2.2 0.9 (0.6–4.6) 0.69 61 D PB1 2.1 1.0 (0.2–4.2) 0.65 61 D

2003 (June–July) (P = 201 mm) MHW 2.7 1.2 (0.6–5.8) 0.81 58 E MRP 2.6 0.5 (0.1–4.7) 0.78 61 E IHW 3.0 1.6 (0.6–6.1) 0.90 57 E IRP 2.7 1.1 (0.4–5.5) 0.82 61 E PB2 2.1 0.9 (0.9–4.2) 0.64 61 F

a Different letter in the same group indicates significant difference at the 0.05 level by ANOVA. Significance tests were performed for the first two groups in column 1 with data for the period from June through September only. season-to-season variation was more pronounced in these Water recharged the soil mostly during the fall months stands than others. (September–October) when precipitation exceeded ET. Annual Modeled daily ET followed the seasonal patterns and total ET for 2003 was similar to that for 2002, but predicted ET measured ET trend well, especially during the peak ET periods for August 2003 was 9–10 mm higher than that for August in July for all sites (Fig. 6). However, the model overestimated 2002, for reasons we have already discussed. Although July PET ET noticeably for MHW, PB1, and PB2 during the spring (May– was higher in 2002 than in 2003, simulated ET for July 2002 was June) in both 2002 and 2003, and in August 2003 for the MRP close to that for July 2003 due to soil moisture stress to tree and PB2 (Fig. 6). For both MRP and MHW, the model transpiration in July 2002. Average annual ET for the mature overestimated the measured ET on the monthly average red pine ecosystem was highest (ranging 575–596 mm/year or basis. However, simulated monthly average ET corresponded 80–86% of precipitation), followed by that for mature hard- more closely to ET measured for PB1 in 2002 and PB2 in 2003 woods (552–554 mm/year or 76–80% of precipitation) and that than to measured ET for the mature forests (Fig. 7). for the clearcut or pine barren system (462–466 mm/year or 64– By the model assumption, when soil moisture is below field 67% of precipitation). capacity, water stress would occur, and modeled actual ET would be less than PET. Our modeling identified at least one period of low ET each year due to soil moisture stress at the 4. Discussion study sites when actual ET was lower than PET, and several stress periods for the low leaf area case. However, the model The forest hydrology of Northern Wisconsin is largely missed one obvious stress period (August 2002) as indicated by controlled by the regional climatic characteristics and the eddy covariance measurements in the mature forests. The dynamics of water and energy inputs during the growing model tracked ET relatively better at the less-vegetated site season from May to October. Forest management practices or (PB1 and PB2) where ET was mostly controlled by soil moisture other land disturbances (e.g., fire and diseases) associated through soil evaporation, which dominated the ET processes with climate changes that alter water loss through the for those two sites. Therefore, we suggest that vapor pressure evapotranspiration process are likely to have profound impact deficit (VPD) might have contributed to low ET during the on the water and energy balances in the region. Knowledge of observed dry periods for forested conditions with high leaf the effects of management practices on energy and water area index values (Fig. 6a). flows in dominant ecosystems in the Upper Great Lakes region The MIKE SHE model provided a complete year-round ET is very limited. This study contrasts total ET in selected data set representing three types of managed stands: mature individual ecosystems with contrasting tree species and tree hardwoods, mature red pine, and a low-LAI stand representing age classes. the pine barren ecosystems (Fig. 8). Monthly ET was the Daily ET varied greatly over the seasons, with rapid highest during the summer months (June–August) (Fig. 8). increase in May, peak in July, and decrease by the end of Water deficits occurred in July as a result of high PET demand. October. Simulation runs with MIKE SHE suggest that ET was 264 agricultural and forest meteorology 148 (2008) 257–267

Fig. 5 – Comparison of monthly ET measured by the eddy covariance method for: (a) mature hardwood (MHW), mature red pine (MRP), recently clearcut hardwoods (YHW), young red pine (YRP), and pine barrens (PB1) during 2002; (b) mature hardwood (MHW), mature red pine (MRP), intermediate red pine (IRP), intermediate hardwoods (YHW), and burned pine barrens (PB2) during 2003.

limited by soil moisture availability in the summer. However, except in the PB1 and PB2 ecosystems, direct flux measurements did not show ET reduction. These discrepancies suggested that the hydrology-based model did not reflect Fig. 6 – Comparison between simulated and measured the physiological control on mature tree transpiration. Recent daily ET for: (a) MRP ecosystem, (b) MHW ecosystem, and studies using the sapflow and eddy flux methods at individual (c) PB1 and PB2 ecosystems during 2002–2003. tree and ecosystem scale have shown clearly that VPD was the key environmental variable controlling forest transpiration through stomatal conductance in the Northern Wisconsin region (Ewers et al., 2002; Mackay et al., 2002). The MIKE SHE strongly influence ecosystem productivity, water availability, hydrologic model is based on physical environmental pro- and water yield in the region. Any changes in the precipitation cesses and does not account for physiological effects, regime due to climatic variability and any changes in the ET including the potential effect of VPD on transpiration, and fluxes due to land management will have more pronounced this may have compromised model predictions. Ecosystem effects on water availability, ecosystem productivity, and modeling indeed suggests that soil moisture and VPD are water yield here than in regions such as the Eastern USA, important in carbon cycling, and thus ET at the study sites in where precipitation is distributed more uniformly throughout general (Ryu et al., this issue). the year (Sun et al., 2004). Seasonal ET changes observed during this study followed The amount of ET loss from one type of forest is largely closely the long-term seasonal distribution of precipitation in controlled by leaf area (Gholz and Clark, 2002; Chen et al., the region. On average, over 70% of annual precipitation in the 2004). Our field data and simulations also confirmed that region falls during the May–October growing season, and this mature forests with higher leaf area had higher ET in the ratio was as high as 78 and 85% in 2002 and 2003, respectively. growing season than those with lower leaf area, such as The timing of water input (precipitation) and output (ET) clearcut ecosystems and pine barrens. However, care must be agricultural and forest meteorology 148 (2008) 257–267 265

recovery of forest transpiration capacity. Also our finding are in good agreement with Verry (1986) who suggests that it takes 12–15 years for harvested aspen stands to recover their water yield during regeneration in the Lake States. In Minnesota, clearcutting of aspen and pine forests increased annual streamflow by 90 and 160 mm/year, respectively, resulting in a 30–80% increase in streamflow or recharge to groundwater (Verry, 1986). However, no detailed specific stand-level data were available to validate these differential responses (Verry, 1986). Our June and July data showed that mature forests lost 5–21% more water (30 mm) in 2002 and 22–42% more water (28–64 mm) in 2003 than ecosystems with low LAI (e.g., PB1, PB2, or YHW). These data provide direct evidence that mature forests use more water than other forest stands during peak Fig. 7 – Comparison of computer simulated and field seasons. In addition, our simulations indicated that convert- measured monthly averaged ET of MRP, MHW, and PB1 ing mature hardwoods or mature red pine to young regener- (2002) and PB2 (2003) (represented as clearcut). A general ating stands would reduce annual ET by 90 and 121 mm, 21% over-prediction of the simulation model is shown. respectively—a range close to values reported by Verry (1986). This water loss is significant in terms of local streamflow and water supply, although the amount of water lost is much smaller than the hydrologic response to forest removal in the taken when predicting ET at the landscape using leaf area as Eastern USA, which is often 200–400 cm/year (Ice and one indictor. Less known are the changes of each of the ET Stednick, 2004). components (transpiration, canopy interception, and soil We found noticeable differences between modeled daily ET evaporation) during forest regeneration. Our data also as predicted by the hydrologic model, and measured daily ET indicated that LAI alone could not explain ET differences as determined by the eddy covariance method. Estimation among different ecosystems with contrasting plant commu- errors were expected for both methods. The eddy covariance nity types. For example, the pine barrens had low LAI but had method does not provide tight energy closure consistently, ET characteristics similar to those of the mature forests and energy closure is very relevant to evaluation of latent heat (Figs. 4a and 5a). Our field data suggested that there were no flux (Wilson et al., 2002) and thus estimation of ET and water significant differences between ET values for the hardwood balance. Wilson et al. (2002) reported an average energy and the coniferous forest during the growing seasons. imbalance of 20% when comparing the sum of LE and sensible However, the model simulations suggested that ET of the heat with available energy (net radiation less energy stored) in red pine forest was 6% higher than that of the hardwoods, a recent review of 22 sites and 50-site years in FLUXNET, a presumably because the former had higher canopy intercep- network of eddy covariance sites. We expected to find tion and transpiration during the non-growing seasons when measurement errors of similar magnitude in our study. Error deciduous trees drop their leaves (Swank and Douglass, 1974). in model prediction may result from inaccuracy of model It appeared that both conifers and deciduous forests had parameterization of soils and groundwater systems and from recovered their ET capacity at age 8–25 years (Figs. 4b and 5b). inaccuracy in simulating the physiological controls (e.g., the This is consistent with findings of Noormets et al. (2007), who effect of VPD on transpiration processes). reported that stands became carbon sinks by age 10 suggesting

5. Conclusion

This study compared ET estimates based on field eddy covariance measurements over two growing seasons (May– October) with computer simulation modeling results in eight dominant ecosystems in a managed landscape in Northern Wisconsin. Comparisons of field data and modeling results were useful in quantifying ET flux at multiple temporal scales, especially as the accuracy of field sampling techniques and models in landscape-scale studies is still limited. We conclude that: (1) prescribed burning that dramatically reduces leaf area reduced ET significantly in the growing seasons; (2) leaf area is not a good indicator of landscape-scale ET; (3) ecosystem ET of regenerated forests recovers from the effects of harvesting by Fig. 8 – Simulated monthly ET (mm) distributions during forest age 8–25 years; (4) overall, simulated ET was higher than 2002–2003 in three contrasting ecosystems, mature red values obtained by the eddy covariance method. The hydro- pine (MRP), mature hardwood (MHW), and clearcut logical model, MIKE SHE, was mostly based on the physical representing pine barrens. processes of water cycles of a forest stand. Although it is still 266 agricultural and forest meteorology 148 (2008) 257–267

useful for identifying the general trend of ET, it lacks Ewers, B.E., Mackay, D.S., Gower, S.T., Ahl, D.E., Burrows, S.N., consideration of the effects of VPD and stomatal activity on Samanta, S., 2002. Tree species effects on stand transpiration in northern Wisconsin. Water Res. Res. 38 (7), tree transpiration. This shortcoming might have caused the ET 1103, doi:10.1029/2001WR000830. overestimation when comparing to the eddy covariance Federer, C.A., Lash, D., 1978. BROOK: A hydrologic simulation method. Future ecosystem-scale independent water balance model for eastern forests. Water Resources Research studies are needed to fully quantify the ET differences among Center, Res. Rep. 19, University of New Hampshire, the major ecosystems in Northern Wisconsin. Durham, NH, 84 pp. Foken, T., Wichura, B., 1996. Tools for quality assessment of surface-based flux measurements. Agric. Forest Meteorol. Acknowledgments 78, 83–105. Gerbert, W.W., Graczkyk, D.J., Krug, W.R., 1987. Average annual runoff in the United States, 1951–1980. In: Hydrol Invest This study was supported by the National Science Foundation Atlas HA-710, USGS, Reston, VA. (DEB0-129405), the Southern Global Change Programs of the Gholz, H.L., Clark, K.L., 2002. Energy exchange across a USDA Forest Service, and the University of Toledo. The chronosequence of slash pine forests in Florida. Agric. authors are grateful to the Washburn Ranger District and Forest Meteorol. 112, 87–102. staff of University of Wisconsin Agricultural Research Station Ice, G.G., Stednick, J.D., 2004. A Century of Forest and Wildland Watershed Lessons. Society of American Foresters, in Ashland for data sharing and providing logistic support. Bethesda, MD, p. 287. Sincere thanks go to Amanda Mermer, John Erhardt, and Im, S., Kim, H., Kim, C., Jang, C., 2004. Evaluation of MIKE SHE Soung-Ryoul Ryu for assistance with data collection and to modeling system to forested watersheds in Korea. ASAE/ Dan Wozniczka and Leif Williams for maintaining field CSAE Meeting Paper No. 042104, St. Joseph, MI. instruments. Three anonymous reviewers, Alan Salmon, Law, B.E., Falge, E., Gu, L., Baldocchi, D.D., Bakwin, P., Berbigier, and a few members of LEES Lab also provided valuable P., Davis, K., Dolman, A.J., Falk, M., Fuentes, J.D., Goldstein, A., Granier, A., Grelle, A., Hollinger, D., Janssens, I.A., Jarvis, suggestions for improving the earlier manuscript. P., Jensen, N.O., Katul, G., Mahli, Y., Matteucci, G., Meyers, T., Monson, R., Munger, W., Oechel, W., Olson, R., Pilegaard, references K., Paw, K.T., Thorgeirsson, H., Valentini, R., Verma, S., Vesala, T., Wilson, K., Wofsy, S., 2002. Environmental controls over carbon dioxide and water vapor exchange of terrestrial vegetation. Agric. Forest Meteorol. 113, 97–120. Abbott, M.B., Bathurst, J.C., Cunge, J.A., O’Connell, P.E., Leuning, R., 2004. Measurements of trace gas fluxes in the Rasmussen, J., 1986a. An introduction to the European atmosphere using eddy covariance: WPL corrections Hydrological System-Systeme Hydrologique Europeen, revisited. In: Lee, X., Massman, W.J., Law, B.E. (Eds.), ‘‘SHE’’. 1. History and philosophy of a physically based, Handbook of Micrometeorology. A Guide to Surface Flux distributed modeling system. J. Hydrol. 87, 45–59. Measurement and Analysis. Kluwer Academic Publishers, Abbott, M.B., Bathurst, J.C., Cunge, J.A., O’Connell, P.E., Boston, MA, USA, pp. 119–132. Rasmussen, J., 1986b. An introduction to the European Lu, J., Sun, G., Amatya, D.M., McNulty, S.G., 2005. Evaluating six Hydrological System-Systeme Hydrologique Europeen SHE. potential evapotranspiration methods in the Southeastern 2. Structure of a physically based, distributed modelling United States. J. Am. Water Res. Assoc. 41, 621–633. system. J. Hydrol. 87, 61–77. Mackay, D.S., Ahl, D.E., Ewers, B.E., Gower, S.T., Burrows, S.N., Andreassian, V., 2004. Waters and forests: from historical Samanta, S., Davis, K.J., 2002. Effects of aggregated controversy to scientific debate. J. Hydrol. 291, 1–27. classifications of forest composition on estimates of Baldocchi, D., Valentini, R., Running, S., Oechel, W.C., Dahlman, evapotranspiration in a northern Wisconsin forest. Global R., 1996. Strategies for measuring and modeling carbon Change Biol. 8 (12), 1253–1265. dioxide and water vapor fluxes over terrestrial ecosystems. McNulty, S.G., Vose, J.M., Swank, W.T., 1997. Regional Global Change Biol. 2, 159–168. hydrologic response of southern pine forests to potential air Bliss, C.M., Comerford, N.B., 2002. Forest harvesting influence temperature and precipitation changes. Water Res. Assoc. on water table dynamics in a Florida flatwoods landscape. Bull. 33, 1011–1022. Soil Soc. Am. J. 66, 1344–1349. Noormets, A., Chen, J., Crow, T.R., in press. Age-dependent Bresee, M.K., Le Moine, J.M., Mather, S., Brosofske, K.D., Chen, J., changes in ecosystem carbon fluxes in managed forests in Crow, T.R., Radmacher, J., 2004. Disturbance and land northern Wisconsin, USA. Ecosystems. dynamics in the Chequamegon National Forest, Wisconsin, Pielke, R.A., Avissar, R., Raupach, M., Dolman, H., Zeng, X., USA, from 1972 to 2001. Lands. Ecol. 19, 291–309. Denning, S., 1998. Interactions between the atmosphere and Chen, J., Falk, M., Euskirchen, E., Paw, U.K.T., Suchanek, T.H., terrestrial ecosystems: influence on weather and climate. Ustin, S.L., Bond, B.J., Brosofske, K., Phillips, N., Bi, R., 2002. Global Change Biol. 4, 461–475. Biophysical controls of carbon flows in three successional Ryu, S., Chen, J., Noormets, A., Ollinger, S., this issue. Douglas-fir stands based on eddy-covariance Comparisons and insights of predicted (PnET-Day) and measurements. Tree Physiol. 22, 169–177. measured gross carbon exchange in Northern Wisconsin Chen, J., Paw, U.K.T., Ustin, S., Suchanek, T., Bond, B.J., Forests. Agric. Forest Meteorol. Brosofske, K.D., Falk, M., 2004. Net ecosystem exchanges of Schafer, K.V., Oren, R., Lai, C., Katul, G.G., 2002. Hydrologic carbon, water, and energy in young and old-growth balance in an intact temperate forest ecosystem under

Douglas-ﬁr forests. Ecosystems 5, 534–544. ambient and elevated atmospheric CO2 concentration. Daly, C., Taylor, G.H., Gibson, W.P., Parzybok, T.W., Johnson, Global Change Biol. 8, 895–911. G.L., Pasteris, P.A., 2000. High-quality spatial climate data Schotanus, P., Nieuwstadt, F.T.M., de Bruin, H.A.R., 1983. sets for the United States and beyond. Am. Soc. Agric. Eng. Temperature measurement with a sonic anemometer and 43 (6), 1957–1962. its application to heat and moisture ﬂuxes. Bound. Lay. DHI, 2004. MIKE SHE User Guide. DHI, Denmark, p. 194. Meteorol. 26, 81–93. agricultural and forest meteorology 148 (2008) 257–267 267

Sun, G., Riekerk, H., Comerford, N.B., 1998. Modeling the and water budgets in evergreen needleaf forests. Agric. hydrologic impacts of forest harvesting on flatwoods. J. Am. Forest Meteorol. 113, 185–222. Water Res. Assoc. 34, 843–854. Van Genuchten, M.T., Leij, F.J., Yates, S.R., 1991. The RETC Code Sun, G., Amatya, D.M., McNulty, S.G., Skaggs, R.W., Hughes, J.H., for Quantifying the Hydraulic Functions of Unsaturated 2000. Climate change impacts on the hydrology and Soils. U.S. Salinity Laboratory, USDA, Riverside, CA, p. 92. productivity of a pine plantation. J. Am. Water Res. Assoc. Verry, E.S., 1986. Forest harvesting and water: the Lake States 36, 367–374. experience. Water Res. Bull. 22, 1039–1047. Sun, G., Riedel, M., Jackson, R., Kolka, R., Amatya, D., Shepard, J. Verry, E.S., 2004. Land fragmentation and impacts to streams 2004. Influences of management of Southern forests on and fish in the central and upper Midwest. In: Ice, G.G., water quantity and quality. In: Rauscher, H.M., Johnsen, K. Stednick, J.D. (Eds.), A Century of Forest and Wildland (Eds.), Southern Forest Sciences: Past, Current, and Future. Watershed Lessons. Society of American Foresters, Gen. Tech. Rep/SRS-75. Department of Agriculture, Forest Bethesda, Maryland, p. 287. Service, Southern Research Station, Ashville, NC, US, p. 394, Webb, E.K., Pearman, G.I., Leuning, R., 1980. Correction of flux Chapter 3. measurements for density effects due to heat and water Swank, W.T., Douglass, J.E., 1974. Streamflow greatly reduced by vapor transfer. Quart. J.R. Meteorol. Soc. 106, 67–90. converting hardwoods to pine. Science 185, 857–859. Werth, D., Avissar, R., 2002. The local and global effects of Swank, W.T., Swift Jr., L.W., Douglass, J.E., 1988. Streamflow Amazon deforestation. J. Geophys. Res. 107, 8087, changes associated with forest cutting, species conversion, doi:10.1029/2001JD000717. and natural disturbances. In: Swank, W.T., Crossley, Jr., D.A. Wilson, K.B., Hanson, P.J., Mulholland, P.J., Baldocchi, D.D., (Eds.), Forest Hydrology and Ecology at Coweeta. Ecological Wullschleger, S.D., 2001. A comparison of methods for Studies, 66. Springer-Verlag, New York, pp. 297–312. determining forest evapotranspiration and its components: Tague, C.L., McMichichael, C., Hope, A., Choate, J., Clark, R., sap-flow, soil water budget, eddy covariance and catchment 2004. Application of the RHESSys model to a California water balance. Agric. Forest Meteorol. 106, 153–168. semiarid shrubland watershed. J. Am. Water Res. Assoc. 40, Wilson, K.A., Goldstein, E., Falge, M., Aubinet, D., Baldocchi, P., 575–589. Berbigier, C., Bernhofer, R., Ceulemans, H., Dolman, C., Thornton, P.E., Law, B.E., Gholz, H.L., Clark, K.L., Falge, E., Field, A., Grelle, A., Ibrom, B.E., Law, A., Kowalski, T., Ellsworth, D.S., Goldstein, A.H., Monson, R.K., Hollinger, D., Meyers, J., Moncrieff, R., Monson, W., Oechel, J., Tenhunen, Falk, M., Chen, J., Sparks, J.P., 2002. Modeling and measuring R., Valentini, S., Verma, 2002. Energy balance closure at the effects of disturbance history and climate on carbon FLUXNET sites. Agric. Forest Meteorol. 113 (1–4), 223–243.